Update app.py

app.py

@@ -7,13 +7,14 @@ from fastapi import FastAPI
 from pydantic import BaseModel
 from transformers import BertTokenizer, AutoModelForSequenceClassification
 from arabert.preprocess import ArabertPreprocessor
+from tabulate import tabulate  
 
 MODEL_REPO = "kkAsmaa/ChildShield"
 MODEL_NAME = "aubmindlab/bert-base-arabertv02-twitter"
 SUB_FOLDER = "ChildShield"
 HF_TOKEN = os.getenv("HF_TOKEN")
 
-print("🔄 Loading ChildShield Model Weights...")
+print("🔄 Loading ChildShield Model Weights with Server-Side Logs Dashboard...")
 tokenizer = BertTokenizer.from_pretrained(MODEL_NAME)
 model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO, token=HF_TOKEN, subfolder=SUB_FOLDER)
 model.eval()
@@ -70,6 +71,7 @@ def predict_safety_api(text):
     highest_safe_prob = 0.0
     windows_analysis = []
     triggered_windows = []
+    windows_table_data = [] 
 
     for idx, win_ids in enumerate(windows):
         window_text = tokenizer.decode(win_ids, skip_special_tokens=True)
@@ -96,31 +98,44 @@ def predict_safety_api(text):
             "prediction": prediction
         })
         
+    
+        windows_table_data.append([
+            f"Win {idx + 1}",
+            window_text[:45] + "..." if len(window_text) > 45 else window_text,
+            f"{safe_prob * 100:.2f}%",
+            f"{unsafe_prob * 100:.2f}%",
+            f"❌ {prediction}" if prediction == "UNSAFE" else f"🍏 {prediction}"
+        ])
+        
         if unsafe_prob > 0.50:
             is_blocked = True
             highest_unsafe_prob = max(highest_unsafe_prob, unsafe_prob)
             triggered_windows.append(idx + 1)
-        
         else:
             highest_safe_prob = max(highest_safe_prob, safe_prob)    
 
     final_prediction = "UNSAFE" if is_blocked else "SAFE"
     winning_probability = highest_unsafe_prob if is_blocked else highest_safe_prob
-    formatted_confidence= f"{winning_probability *100:.2f}%"
+    formatted_confidence = f"{winning_probability * 100:.2f}%"
     
     unsafe_confidence_score = round(highest_unsafe_prob, 4)
     safe_confidence_score = round(1.0 - highest_unsafe_prob, 4)
     final_confidence = unsafe_confidence_score if is_blocked else safe_confidence_score
 
-    print("\n📊 ===== CHILDSHIELD REPORT =====")
-    print(f"📥 Original Text:\n{text[:100]}")
-    print(f"\n🧹 Cleaned Text:\n{cleaned_text[:100]}")
-    print(f"\n🔑 Total Tokens: {total_tokens_count}")
-    print(f"🪟 Total Windows: {total_windows_count}")
-    print(f"🚨 Final Verdict: {final_prediction}")
-    print(f"🛑 Triggered Windows ID: {triggered_windows}")
-    print(f"🏆 Model Confidence On Winning Class: {formatted_confidence}")
-    print("=================================\n")
+
+    alert_banner = "🚨 [BLOCK] CHILDSHIELD AI INFERENCE REPORT" if is_blocked else "🍏 [PASS] CHILDSHIELD AI INFERENCE REPORT"
+    print(f"\n================ {alert_banner} ================")
+    print(f"📥 Received Original Text:\n\"{text.strip()}\"")
+    print(f"\n🧹 Preprocessed Cleaned Text:\n\"{cleaned_text}\"")
+    print(f"\n🔑 Total Page Tokens Count  : {total_tokens_count}")
+    print(f"🪟 Total Sliding Windows Run : {total_windows_count} Windows (Size: 60, Overlap: 20)")
+    print(f"🎯 Final Security Verdict    : {final_prediction}")
+    print(f"🏆 Model Decision Confidence : {formatted_confidence}")
+    print(f"🛑 Triggered Windows ID      : {triggered_windows}")
+    print("\n📊 --- Windows Detailed Semantic Analysis Table ---")
+   
+    print(tabulate(windows_table_data, headers=["ID", "Window Text Preview", "Safe Prob", "Unsafe Prob", "Verdict"], tablefmt="grid"))
+    print("========================================================================\n")
 
     try:
         log_file_path = "production_logs.txt"
@@ -160,3 +175,4 @@ app = gr.mount_gradio_app(app, gradio_interface, path="/")
 
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=7860)
+